Performance Characterization of the Quad Pentium Pro SMP Using OLTP
Kimberly Keeton, David A. Patterson, Yong Qiang He, Roger C. Raphael, Walter E. Baker · 1999
Commercial applications are an important, yet often overlooked, workload with significantly different characteristics from technical workloads. In this paper, we use hardware counters to measure architectural features of a four-processor Pentium Pro-based server running a commercial database executing a TPC-C-like workload. We examine the efficiency of out-of-order execution, branch prediction, speculative execution, and caching. In addition, we study the sensitivity of database performance to L2 cache size, main memory bandwidth and the number of processors. We find that out-of-order execution, superscalar issue and retire, and branch prediction are less efficient for database workloads than for technical applications. We found that caches are effective at reducing processor traffic to memory, and make the workload more tolerant of memory bandwidth reductions. Larger caches would be helpful to absorb more data requests, however.